What the AI-Driven DRAM Squeeze Means
The AI-driven DRAM squeeze is a market shift where booming AI memory demand for high-bandwidth memory (HBM) pulls manufacturing capacity away from mainstream DDR5 and DDR4 chips, tightening supply and pushing up DRAM prices for everyday PCs and servers. As AI accelerators consume more HBM, top memory manufacturers are reallocating production lines toward these premium products. That leaves fewer wafers and less fab time for standard PC-use DRAM, even while demand for desktops, laptops, and general servers continues. The result is a DDR5 DDR4 shortage that shows up first in contract prices, then in retail shelves and PC building costs. For consumers and system integrators, this dynamic means fewer bargains, more volatile pricing, and a higher baseline cost for memory-intensive builds in the near term.
HBM Memory Production Takes Priority
High-bandwidth memory sits at the heart of the current AI boom. Training and running large AI models requires enormous data throughput, so chipmakers pair AI accelerators with stacks of HBM instead of conventional DRAM. To keep up, Samsung, SK Hynix, and Micron are prioritizing HBM memory production over mainstream DDR5 and DDR4. These firms share the same fabs, equipment, and engineers across product lines, so every wafer committed to HBM is a wafer not used for commodity PC DRAM. At the same time, demand for other specialized memory, such as single-level cell NAND for edge AI and communications networks, is also rising. Suppliers are keeping legacy production capacity low and show no sign of restoring older processes, reinforcing the tilt toward high-margin AI-focused products and away from the bulk PC memory that builders rely on.
DDR5 and DDR4 Shortage Hits DRAM Prices
The shift toward AI memory demand is already visible in DRAM prices, especially for commodity PC-use parts. According to DRAMeXchange data cited by TrendForce, the average price of DDR4 8Gb PC DRAM reached USD 20 (approx. RM92) in May, up from USD 16 (approx. RM73) in April, a 25% jump and the highest level since tracking began in June 2016. TrendForce notes that negotiated PC DRAM contract prices rose 40–50% from the previous quarter, after an even sharper 100–115% surge in the first quarter. NAND flash is on a similar trajectory, with 128Gb MLC products climbing for 17 consecutive months. When mainstream DDR5 and DDR4 production shrinks while demand stays firm, a DRAM price increase is almost unavoidable, and the current AI-first production strategy is amplifying that effect.
Rising PC Building Costs for Consumers and Integrators
For PC builders, the DDR5 DDR4 shortage translates into higher PC building costs and fewer memory options. Contract price spikes for commodity DRAM tend to flow into retail channels with a delay, but system integrators and OEMs are already absorbing higher quotes for standard modules. With DRAM suppliers focused on HBM and keeping overall production constrained, there is little slack to soften the blow. Builders planning new gaming rigs, workstations, or home labs will find that memory now takes a larger share of the budget than in recent years. Storage is also more expensive as NAND prices rise. While other parts—CPUs, GPUs, and cases—may offer occasional discounts, the memory portion of a bill of materials is trending upward, making it harder to deliver low-cost systems without cutting capacity or speed.
How Long Will the DRAM Price Pressure Last?
The key question for buyers is how long AI memory demand will dominate production decisions. As long as cloud providers and chipmakers race to deploy new AI servers, Samsung, SK Hynix, and Micron have strong reasons to prioritize HBM memory production and maintain tight supply for commodity DRAM. TrendForce expects DRAM prices in June to stay near May levels, suggesting a pause rather than a reversal after steep gains earlier this year. Suppliers are also keeping low capacity for mature and legacy products, reinforcing the overall shortage. For PC builders, that means planning for persistently higher DRAM price levels rather than waiting for a quick collapse. Locking in memory purchases earlier in a project and being flexible on exact speeds or capacities can help soften the impact while the market remains constrained.
